2 * DCA compatible decoder
3 * Copyright (C) 2004 Gildas Bazin
4 * Copyright (C) 2004 Benjamin Zores
5 * Copyright (C) 2006 Benjamin Larsson
6 * Copyright (C) 2007 Konstantin Shishkov
8 * This file is part of FFmpeg.
10 * FFmpeg is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU Lesser General Public
12 * License as published by the Free Software Foundation; either
13 * version 2.1 of the License, or (at your option) any later version.
15 * FFmpeg is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * Lesser General Public License for more details.
20 * You should have received a copy of the GNU Lesser General Public
21 * License along with FFmpeg; if not, write to the Free Software
22 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
29 #include "libavutil/channel_layout.h"
30 #include "libavutil/common.h"
31 #include "libavutil/float_dsp.h"
32 #include "libavutil/internal.h"
33 #include "libavutil/intreadwrite.h"
34 #include "libavutil/mathematics.h"
35 #include "libavutil/opt.h"
36 #include "libavutil/samplefmt.h"
40 #include "dca_syncwords.h"
45 #include "fmtconvert.h"
49 #include "synth_filter.h"
70 enum DCAXxchSpeakerMask {
71 DCA_XXCH_FRONT_CENTER = 0x0000001,
72 DCA_XXCH_FRONT_LEFT = 0x0000002,
73 DCA_XXCH_FRONT_RIGHT = 0x0000004,
74 DCA_XXCH_SIDE_REAR_LEFT = 0x0000008,
75 DCA_XXCH_SIDE_REAR_RIGHT = 0x0000010,
76 DCA_XXCH_LFE1 = 0x0000020,
77 DCA_XXCH_REAR_CENTER = 0x0000040,
78 DCA_XXCH_SURROUND_REAR_LEFT = 0x0000080,
79 DCA_XXCH_SURROUND_REAR_RIGHT = 0x0000100,
80 DCA_XXCH_SIDE_SURROUND_LEFT = 0x0000200,
81 DCA_XXCH_SIDE_SURROUND_RIGHT = 0x0000400,
82 DCA_XXCH_FRONT_CENTER_LEFT = 0x0000800,
83 DCA_XXCH_FRONT_CENTER_RIGHT = 0x0001000,
84 DCA_XXCH_FRONT_HIGH_LEFT = 0x0002000,
85 DCA_XXCH_FRONT_HIGH_CENTER = 0x0004000,
86 DCA_XXCH_FRONT_HIGH_RIGHT = 0x0008000,
87 DCA_XXCH_LFE2 = 0x0010000,
88 DCA_XXCH_SIDE_FRONT_LEFT = 0x0020000,
89 DCA_XXCH_SIDE_FRONT_RIGHT = 0x0040000,
90 DCA_XXCH_OVERHEAD = 0x0080000,
91 DCA_XXCH_SIDE_HIGH_LEFT = 0x0100000,
92 DCA_XXCH_SIDE_HIGH_RIGHT = 0x0200000,
93 DCA_XXCH_REAR_HIGH_CENTER = 0x0400000,
94 DCA_XXCH_REAR_HIGH_LEFT = 0x0800000,
95 DCA_XXCH_REAR_HIGH_RIGHT = 0x1000000,
96 DCA_XXCH_REAR_LOW_CENTER = 0x2000000,
97 DCA_XXCH_REAR_LOW_LEFT = 0x4000000,
98 DCA_XXCH_REAR_LOW_RIGHT = 0x8000000,
101 #define DCA_DOLBY 101 /* FIXME */
103 #define DCA_CHANNEL_BITS 6
104 #define DCA_CHANNEL_MASK 0x3F
108 #define HEADER_SIZE 14
110 #define DCA_NSYNCAUX 0x9A1105A0
113 /** Bit allocation */
114 typedef struct BitAlloc {
115 int offset; ///< code values offset
116 int maxbits[8]; ///< max bits in VLC
117 int wrap; ///< wrap for get_vlc2()
118 VLC vlc[8]; ///< actual codes
121 static BitAlloc dca_bitalloc_index; ///< indexes for samples VLC select
122 static BitAlloc dca_tmode; ///< transition mode VLCs
123 static BitAlloc dca_scalefactor; ///< scalefactor VLCs
124 static BitAlloc dca_smpl_bitalloc[11]; ///< samples VLCs
126 static av_always_inline int get_bitalloc(GetBitContext *gb, BitAlloc *ba,
129 return get_vlc2(gb, ba->vlc[idx].table, ba->vlc[idx].bits, ba->wrap) +
133 static float dca_dmix_code(unsigned code);
135 static av_cold void dca_init_vlcs(void)
137 static int vlcs_initialized = 0;
139 static VLC_TYPE dca_table[23622][2];
141 if (vlcs_initialized)
144 dca_bitalloc_index.offset = 1;
145 dca_bitalloc_index.wrap = 2;
146 for (i = 0; i < 5; i++) {
147 dca_bitalloc_index.vlc[i].table = &dca_table[ff_dca_vlc_offs[i]];
148 dca_bitalloc_index.vlc[i].table_allocated = ff_dca_vlc_offs[i + 1] - ff_dca_vlc_offs[i];
149 init_vlc(&dca_bitalloc_index.vlc[i], bitalloc_12_vlc_bits[i], 12,
150 bitalloc_12_bits[i], 1, 1,
151 bitalloc_12_codes[i], 2, 2, INIT_VLC_USE_NEW_STATIC);
153 dca_scalefactor.offset = -64;
154 dca_scalefactor.wrap = 2;
155 for (i = 0; i < 5; i++) {
156 dca_scalefactor.vlc[i].table = &dca_table[ff_dca_vlc_offs[i + 5]];
157 dca_scalefactor.vlc[i].table_allocated = ff_dca_vlc_offs[i + 6] - ff_dca_vlc_offs[i + 5];
158 init_vlc(&dca_scalefactor.vlc[i], SCALES_VLC_BITS, 129,
159 scales_bits[i], 1, 1,
160 scales_codes[i], 2, 2, INIT_VLC_USE_NEW_STATIC);
162 dca_tmode.offset = 0;
164 for (i = 0; i < 4; i++) {
165 dca_tmode.vlc[i].table = &dca_table[ff_dca_vlc_offs[i + 10]];
166 dca_tmode.vlc[i].table_allocated = ff_dca_vlc_offs[i + 11] - ff_dca_vlc_offs[i + 10];
167 init_vlc(&dca_tmode.vlc[i], tmode_vlc_bits[i], 4,
169 tmode_codes[i], 2, 2, INIT_VLC_USE_NEW_STATIC);
172 for (i = 0; i < 10; i++)
173 for (j = 0; j < 7; j++) {
174 if (!bitalloc_codes[i][j])
176 dca_smpl_bitalloc[i + 1].offset = bitalloc_offsets[i];
177 dca_smpl_bitalloc[i + 1].wrap = 1 + (j > 4);
178 dca_smpl_bitalloc[i + 1].vlc[j].table = &dca_table[ff_dca_vlc_offs[c]];
179 dca_smpl_bitalloc[i + 1].vlc[j].table_allocated = ff_dca_vlc_offs[c + 1] - ff_dca_vlc_offs[c];
181 init_vlc(&dca_smpl_bitalloc[i + 1].vlc[j], bitalloc_maxbits[i][j],
183 bitalloc_bits[i][j], 1, 1,
184 bitalloc_codes[i][j], 2, 2, INIT_VLC_USE_NEW_STATIC);
187 vlcs_initialized = 1;
190 static inline void get_array(GetBitContext *gb, int *dst, int len, int bits)
193 *dst++ = get_bits(gb, bits);
196 static inline int dca_xxch2index(DCAContext *s, int xxch_ch)
200 /* locate channel set containing the channel */
201 for (i = -1, base = 0, mask = (s->xxch_core_spkmask & ~DCA_XXCH_LFE1);
202 i <= s->xxch_chset && !(mask & xxch_ch); mask = s->xxch_spk_masks[++i])
203 base += av_popcount(mask);
205 return base + av_popcount(mask & (xxch_ch - 1));
208 static int dca_parse_audio_coding_header(DCAContext *s, int base_channel,
212 static const float adj_table[4] = { 1.0, 1.1250, 1.2500, 1.4375 };
213 static const int bitlen[11] = { 0, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3 };
214 static const int thr[11] = { 0, 1, 3, 3, 3, 3, 7, 7, 7, 7, 7 };
215 int hdr_pos = 0, hdr_size = 0;
217 int this_chans, acc_mask;
218 int embedded_downmix;
222 /* xxch has arbitrary sized audio coding headers */
224 hdr_pos = get_bits_count(&s->gb);
225 hdr_size = get_bits(&s->gb, 7) + 1;
228 nchans = get_bits(&s->gb, 3) + 1;
229 s->total_channels = nchans + base_channel;
230 s->prim_channels = s->total_channels;
232 /* obtain speaker layout mask & downmix coefficients for XXCH */
234 acc_mask = s->xxch_core_spkmask;
236 this_chans = get_bits(&s->gb, s->xxch_nbits_spk_mask - 6) << 6;
237 s->xxch_spk_masks[s->xxch_chset] = this_chans;
238 s->xxch_chset_nch[s->xxch_chset] = nchans;
240 for (i = 0; i <= s->xxch_chset; i++)
241 acc_mask |= s->xxch_spk_masks[i];
243 /* check for downmixing information */
244 if (get_bits1(&s->gb)) {
245 embedded_downmix = get_bits1(&s->gb);
246 coeff = get_bits(&s->gb, 6);
248 if (coeff<1 || coeff>61) {
249 av_log(s->avctx, AV_LOG_ERROR, "6bit coeff %d is out of range\n", coeff);
250 return AVERROR_INVALIDDATA;
253 scale_factor = -1.0f / dca_dmix_code((coeff<<2)-3);
255 s->xxch_dmix_sf[s->xxch_chset] = scale_factor;
257 for (i = base_channel; i < s->prim_channels; i++) {
258 mask[i] = get_bits(&s->gb, s->xxch_nbits_spk_mask);
261 for (j = base_channel; j < s->prim_channels; j++) {
262 memset(s->xxch_dmix_coeff[j], 0, sizeof(s->xxch_dmix_coeff[0]));
263 s->xxch_dmix_embedded |= (embedded_downmix << j);
264 for (i = 0; i < s->xxch_nbits_spk_mask; i++) {
265 if (mask[j] & (1 << i)) {
266 if ((1 << i) == DCA_XXCH_LFE1) {
267 av_log(s->avctx, AV_LOG_WARNING,
268 "DCA-XXCH: dmix to LFE1 not supported.\n");
272 coeff = get_bits(&s->gb, 7);
273 ichan = dca_xxch2index(s, 1 << i);
274 if ((coeff&63)<1 || (coeff&63)>61) {
275 av_log(s->avctx, AV_LOG_ERROR, "7bit coeff %d is out of range\n", coeff);
276 return AVERROR_INVALIDDATA;
278 s->xxch_dmix_coeff[j][ichan] = dca_dmix_code((coeff<<2)-3);
285 if (s->prim_channels > DCA_PRIM_CHANNELS_MAX)
286 s->prim_channels = DCA_PRIM_CHANNELS_MAX;
288 for (i = base_channel; i < s->prim_channels; i++) {
289 s->subband_activity[i] = get_bits(&s->gb, 5) + 2;
290 if (s->subband_activity[i] > DCA_SUBBANDS)
291 s->subband_activity[i] = DCA_SUBBANDS;
293 for (i = base_channel; i < s->prim_channels; i++) {
294 s->vq_start_subband[i] = get_bits(&s->gb, 5) + 1;
295 if (s->vq_start_subband[i] > DCA_SUBBANDS)
296 s->vq_start_subband[i] = DCA_SUBBANDS;
298 get_array(&s->gb, s->joint_intensity + base_channel, s->prim_channels - base_channel, 3);
299 get_array(&s->gb, s->transient_huffman + base_channel, s->prim_channels - base_channel, 2);
300 get_array(&s->gb, s->scalefactor_huffman + base_channel, s->prim_channels - base_channel, 3);
301 get_array(&s->gb, s->bitalloc_huffman + base_channel, s->prim_channels - base_channel, 3);
303 /* Get codebooks quantization indexes */
305 memset(s->quant_index_huffman, 0, sizeof(s->quant_index_huffman));
306 for (j = 1; j < 11; j++)
307 for (i = base_channel; i < s->prim_channels; i++)
308 s->quant_index_huffman[i][j] = get_bits(&s->gb, bitlen[j]);
310 /* Get scale factor adjustment */
311 for (j = 0; j < 11; j++)
312 for (i = base_channel; i < s->prim_channels; i++)
313 s->scalefactor_adj[i][j] = 1;
315 for (j = 1; j < 11; j++)
316 for (i = base_channel; i < s->prim_channels; i++)
317 if (s->quant_index_huffman[i][j] < thr[j])
318 s->scalefactor_adj[i][j] = adj_table[get_bits(&s->gb, 2)];
321 if (s->crc_present) {
322 /* Audio header CRC check */
323 get_bits(&s->gb, 16);
326 /* Skip to the end of the header, also ignore CRC if present */
327 i = get_bits_count(&s->gb);
328 if (hdr_pos + 8 * hdr_size > i)
329 skip_bits_long(&s->gb, hdr_pos + 8 * hdr_size - i);
332 s->current_subframe = 0;
333 s->current_subsubframe = 0;
338 static int dca_parse_frame_header(DCAContext *s)
340 init_get_bits(&s->gb, s->dca_buffer, s->dca_buffer_size * 8);
343 skip_bits_long(&s->gb, 32);
346 s->frame_type = get_bits(&s->gb, 1);
347 s->samples_deficit = get_bits(&s->gb, 5) + 1;
348 s->crc_present = get_bits(&s->gb, 1);
349 s->sample_blocks = get_bits(&s->gb, 7) + 1;
350 s->frame_size = get_bits(&s->gb, 14) + 1;
351 if (s->frame_size < 95)
352 return AVERROR_INVALIDDATA;
353 s->amode = get_bits(&s->gb, 6);
354 s->sample_rate = avpriv_dca_sample_rates[get_bits(&s->gb, 4)];
356 return AVERROR_INVALIDDATA;
357 s->bit_rate_index = get_bits(&s->gb, 5);
358 s->bit_rate = ff_dca_bit_rates[s->bit_rate_index];
360 return AVERROR_INVALIDDATA;
362 skip_bits1(&s->gb); // always 0 (reserved, cf. ETSI TS 102 114 V1.4.1)
363 s->dynrange = get_bits(&s->gb, 1);
364 s->timestamp = get_bits(&s->gb, 1);
365 s->aux_data = get_bits(&s->gb, 1);
366 s->hdcd = get_bits(&s->gb, 1);
367 s->ext_descr = get_bits(&s->gb, 3);
368 s->ext_coding = get_bits(&s->gb, 1);
369 s->aspf = get_bits(&s->gb, 1);
370 s->lfe = get_bits(&s->gb, 2);
371 s->predictor_history = get_bits(&s->gb, 1);
375 av_log(s->avctx, AV_LOG_ERROR, "Invalid LFE value: %d\n", s->lfe);
376 return AVERROR_INVALIDDATA;
379 /* TODO: check CRC */
381 s->header_crc = get_bits(&s->gb, 16);
383 s->multirate_inter = get_bits(&s->gb, 1);
384 s->version = get_bits(&s->gb, 4);
385 s->copy_history = get_bits(&s->gb, 2);
386 s->source_pcm_res = get_bits(&s->gb, 3);
387 s->front_sum = get_bits(&s->gb, 1);
388 s->surround_sum = get_bits(&s->gb, 1);
389 s->dialog_norm = get_bits(&s->gb, 4);
391 /* FIXME: channels mixing levels */
392 s->output = s->amode;
394 s->output |= DCA_LFE;
396 /* Primary audio coding header */
397 s->subframes = get_bits(&s->gb, 4) + 1;
399 return dca_parse_audio_coding_header(s, 0, 0);
402 static inline int get_scale(GetBitContext *gb, int level, int value, int log2range)
405 /* huffman encoded */
406 value += get_bitalloc(gb, &dca_scalefactor, level);
407 value = av_clip(value, 0, (1 << log2range) - 1);
408 } else if (level < 8) {
409 if (level + 1 > log2range) {
410 skip_bits(gb, level + 1 - log2range);
411 value = get_bits(gb, log2range);
413 value = get_bits(gb, level + 1);
419 static int dca_subframe_header(DCAContext *s, int base_channel, int block_index)
421 /* Primary audio coding side information */
424 if (get_bits_left(&s->gb) < 0)
425 return AVERROR_INVALIDDATA;
428 s->subsubframes[s->current_subframe] = get_bits(&s->gb, 2) + 1;
429 s->partial_samples[s->current_subframe] = get_bits(&s->gb, 3);
432 for (j = base_channel; j < s->prim_channels; j++) {
433 for (k = 0; k < s->subband_activity[j]; k++)
434 s->prediction_mode[j][k] = get_bits(&s->gb, 1);
437 /* Get prediction codebook */
438 for (j = base_channel; j < s->prim_channels; j++) {
439 for (k = 0; k < s->subband_activity[j]; k++) {
440 if (s->prediction_mode[j][k] > 0) {
441 /* (Prediction coefficient VQ address) */
442 s->prediction_vq[j][k] = get_bits(&s->gb, 12);
447 /* Bit allocation index */
448 for (j = base_channel; j < s->prim_channels; j++) {
449 for (k = 0; k < s->vq_start_subband[j]; k++) {
450 if (s->bitalloc_huffman[j] == 6)
451 s->bitalloc[j][k] = get_bits(&s->gb, 5);
452 else if (s->bitalloc_huffman[j] == 5)
453 s->bitalloc[j][k] = get_bits(&s->gb, 4);
454 else if (s->bitalloc_huffman[j] == 7) {
455 av_log(s->avctx, AV_LOG_ERROR,
456 "Invalid bit allocation index\n");
457 return AVERROR_INVALIDDATA;
460 get_bitalloc(&s->gb, &dca_bitalloc_index, s->bitalloc_huffman[j]);
463 if (s->bitalloc[j][k] > 26) {
464 av_dlog(s->avctx, "bitalloc index [%i][%i] too big (%i)\n",
465 j, k, s->bitalloc[j][k]);
466 return AVERROR_INVALIDDATA;
471 /* Transition mode */
472 for (j = base_channel; j < s->prim_channels; j++) {
473 for (k = 0; k < s->subband_activity[j]; k++) {
474 s->transition_mode[j][k] = 0;
475 if (s->subsubframes[s->current_subframe] > 1 &&
476 k < s->vq_start_subband[j] && s->bitalloc[j][k] > 0) {
477 s->transition_mode[j][k] =
478 get_bitalloc(&s->gb, &dca_tmode, s->transient_huffman[j]);
483 if (get_bits_left(&s->gb) < 0)
484 return AVERROR_INVALIDDATA;
486 for (j = base_channel; j < s->prim_channels; j++) {
487 const uint32_t *scale_table;
488 int scale_sum, log_size;
490 memset(s->scale_factor[j], 0,
491 s->subband_activity[j] * sizeof(s->scale_factor[0][0][0]) * 2);
493 if (s->scalefactor_huffman[j] == 6) {
494 scale_table = ff_dca_scale_factor_quant7;
497 scale_table = ff_dca_scale_factor_quant6;
501 /* When huffman coded, only the difference is encoded */
504 for (k = 0; k < s->subband_activity[j]; k++) {
505 if (k >= s->vq_start_subband[j] || s->bitalloc[j][k] > 0) {
506 scale_sum = get_scale(&s->gb, s->scalefactor_huffman[j], scale_sum, log_size);
507 s->scale_factor[j][k][0] = scale_table[scale_sum];
510 if (k < s->vq_start_subband[j] && s->transition_mode[j][k]) {
511 /* Get second scale factor */
512 scale_sum = get_scale(&s->gb, s->scalefactor_huffman[j], scale_sum, log_size);
513 s->scale_factor[j][k][1] = scale_table[scale_sum];
518 /* Joint subband scale factor codebook select */
519 for (j = base_channel; j < s->prim_channels; j++) {
520 /* Transmitted only if joint subband coding enabled */
521 if (s->joint_intensity[j] > 0)
522 s->joint_huff[j] = get_bits(&s->gb, 3);
525 if (get_bits_left(&s->gb) < 0)
526 return AVERROR_INVALIDDATA;
528 /* Scale factors for joint subband coding */
529 for (j = base_channel; j < s->prim_channels; j++) {
532 /* Transmitted only if joint subband coding enabled */
533 if (s->joint_intensity[j] > 0) {
535 source_channel = s->joint_intensity[j] - 1;
537 /* When huffman coded, only the difference is encoded
538 * (is this valid as well for joint scales ???) */
540 for (k = s->subband_activity[j]; k < s->subband_activity[source_channel]; k++) {
541 scale = get_scale(&s->gb, s->joint_huff[j], 64 /* bias */, 7);
542 s->joint_scale_factor[j][k] = scale; /*joint_scale_table[scale]; */
545 if (!(s->debug_flag & 0x02)) {
546 av_log(s->avctx, AV_LOG_DEBUG,
547 "Joint stereo coding not supported\n");
548 s->debug_flag |= 0x02;
553 /* Dynamic range coefficient */
554 if (!base_channel && s->dynrange)
555 s->dynrange_coef = get_bits(&s->gb, 8);
557 /* Side information CRC check word */
558 if (s->crc_present) {
559 get_bits(&s->gb, 16);
563 * Primary audio data arrays
566 /* VQ encoded high frequency subbands */
567 for (j = base_channel; j < s->prim_channels; j++)
568 for (k = s->vq_start_subband[j]; k < s->subband_activity[j]; k++)
569 /* 1 vector -> 32 samples */
570 s->high_freq_vq[j][k] = get_bits(&s->gb, 10);
572 /* Low frequency effect data */
573 if (!base_channel && s->lfe) {
576 int lfe_samples = 2 * s->lfe * (4 + block_index);
577 int lfe_end_sample = 2 * s->lfe * (4 + block_index + s->subsubframes[s->current_subframe]);
580 for (j = lfe_samples; j < lfe_end_sample; j++) {
581 /* Signed 8 bits int */
582 s->lfe_data[j] = get_sbits(&s->gb, 8);
585 /* Scale factor index */
586 quant7 = get_bits(&s->gb, 8);
588 avpriv_request_sample(s->avctx, "LFEScaleIndex larger than 127");
589 return AVERROR_INVALIDDATA;
591 s->lfe_scale_factor = ff_dca_scale_factor_quant7[quant7];
593 /* Quantization step size * scale factor */
594 lfe_scale = 0.035 * s->lfe_scale_factor;
596 for (j = lfe_samples; j < lfe_end_sample; j++)
597 s->lfe_data[j] *= lfe_scale;
603 static void qmf_32_subbands(DCAContext *s, int chans,
604 float samples_in[32][8], float *samples_out,
607 const float *prCoeff;
609 int sb_act = s->subband_activity[chans];
611 scale *= sqrt(1 / 8.0);
614 if (!s->multirate_inter) /* Non-perfect reconstruction */
615 prCoeff = ff_dca_fir_32bands_nonperfect;
616 else /* Perfect reconstruction */
617 prCoeff = ff_dca_fir_32bands_perfect;
619 s->dcadsp.qmf_32_subbands(samples_in, sb_act, &s->synth, &s->imdct,
620 s->subband_fir_hist[chans],
621 &s->hist_index[chans],
622 s->subband_fir_noidea[chans], prCoeff,
623 samples_out, s->raXin, scale);
626 static void lfe_interpolation_fir(DCAContext *s, int decimation_select,
627 int num_deci_sample, float *samples_in,
630 /* samples_in: An array holding decimated samples.
631 * Samples in current subframe starts from samples_in[0],
632 * while samples_in[-1], samples_in[-2], ..., stores samples
633 * from last subframe as history.
635 * samples_out: An array holding interpolated samples
639 const float *prCoeff;
642 /* Select decimation filter */
643 if (decimation_select == 1) {
645 prCoeff = ff_dca_lfe_fir_128;
648 prCoeff = ff_dca_lfe_fir_64;
651 for (deciindex = 0; deciindex < num_deci_sample; deciindex++) {
652 s->dcadsp.lfe_fir[idx](samples_out, samples_in, prCoeff);
654 samples_out += 2 * 32 * (1 + idx);
658 /* downmixing routines */
659 #define MIX_REAR1(samples, s1, rs, coef) \
660 samples[0][i] += samples[s1][i] * coef[rs][0]; \
661 samples[1][i] += samples[s1][i] * coef[rs][1];
663 #define MIX_REAR2(samples, s1, s2, rs, coef) \
664 samples[0][i] += samples[s1][i] * coef[rs][0] + samples[s2][i] * coef[rs + 1][0]; \
665 samples[1][i] += samples[s1][i] * coef[rs][1] + samples[s2][i] * coef[rs + 1][1];
667 #define MIX_FRONT3(samples, coef) \
671 samples[0][i] = t * coef[0][0] + u * coef[1][0] + v * coef[2][0]; \
672 samples[1][i] = t * coef[0][1] + u * coef[1][1] + v * coef[2][1];
674 #define DOWNMIX_TO_STEREO(op1, op2) \
675 for (i = 0; i < 256; i++) { \
680 static void dca_downmix(float **samples, int srcfmt, int lfe_present,
681 float coef[DCA_PRIM_CHANNELS_MAX + 1][2],
682 const int8_t *channel_mapping)
684 int c, l, r, sl, sr, s;
691 av_log(NULL, AV_LOG_ERROR, "Not implemented!\n");
695 case DCA_STEREO_TOTAL:
696 case DCA_STEREO_SUMDIFF:
699 c = channel_mapping[0];
700 l = channel_mapping[1];
701 r = channel_mapping[2];
702 DOWNMIX_TO_STEREO(MIX_FRONT3(samples, coef), );
705 s = channel_mapping[2];
706 DOWNMIX_TO_STEREO(MIX_REAR1(samples, s, 2, coef), );
709 c = channel_mapping[0];
710 l = channel_mapping[1];
711 r = channel_mapping[2];
712 s = channel_mapping[3];
713 DOWNMIX_TO_STEREO(MIX_FRONT3(samples, coef),
714 MIX_REAR1(samples, s, 3, coef));
717 sl = channel_mapping[2];
718 sr = channel_mapping[3];
719 DOWNMIX_TO_STEREO(MIX_REAR2(samples, sl, sr, 2, coef), );
722 c = channel_mapping[0];
723 l = channel_mapping[1];
724 r = channel_mapping[2];
725 sl = channel_mapping[3];
726 sr = channel_mapping[4];
727 DOWNMIX_TO_STEREO(MIX_FRONT3(samples, coef),
728 MIX_REAR2(samples, sl, sr, 3, coef));
732 int lf_buf = ff_dca_lfe_index[srcfmt];
733 int lf_idx = ff_dca_channels[srcfmt];
734 for (i = 0; i < 256; i++) {
735 samples[0][i] += samples[lf_buf][i] * coef[lf_idx][0];
736 samples[1][i] += samples[lf_buf][i] * coef[lf_idx][1];
741 #ifndef decode_blockcodes
742 /* Very compact version of the block code decoder that does not use table
743 * look-up but is slightly slower */
744 static int decode_blockcode(int code, int levels, int32_t *values)
747 int offset = (levels - 1) >> 1;
749 for (i = 0; i < 4; i++) {
750 int div = FASTDIV(code, levels);
751 values[i] = code - offset - div * levels;
758 static int decode_blockcodes(int code1, int code2, int levels, int32_t *values)
760 return decode_blockcode(code1, levels, values) |
761 decode_blockcode(code2, levels, values + 4);
765 static const uint8_t abits_sizes[7] = { 7, 10, 12, 13, 15, 17, 19 };
766 static const uint8_t abits_levels[7] = { 3, 5, 7, 9, 13, 17, 25 };
768 static int dca_subsubframe(DCAContext *s, int base_channel, int block_index)
771 int subsubframe = s->current_subsubframe;
773 const float *quant_step_table;
776 float (*subband_samples)[DCA_SUBBANDS][8] = s->subband_samples[block_index];
777 LOCAL_ALIGNED_16(int32_t, block, [8 * DCA_SUBBANDS]);
783 /* Select quantization step size table */
784 if (s->bit_rate_index == 0x1f)
785 quant_step_table = ff_dca_lossless_quant_d;
787 quant_step_table = ff_dca_lossy_quant_d;
789 for (k = base_channel; k < s->prim_channels; k++) {
790 float rscale[DCA_SUBBANDS];
792 if (get_bits_left(&s->gb) < 0)
793 return AVERROR_INVALIDDATA;
795 for (l = 0; l < s->vq_start_subband[k]; l++) {
798 /* Select the mid-tread linear quantizer */
799 int abits = s->bitalloc[k][l];
801 float quant_step_size = quant_step_table[abits];
804 * Determine quantization index code book and its type
807 /* Select quantization index code book */
808 int sel = s->quant_index_huffman[k][abits];
811 * Extract bits from the bit stream
815 memset(block + 8 * l, 0, 8 * sizeof(block[0]));
817 /* Deal with transients */
818 int sfi = s->transition_mode[k][l] && subsubframe >= s->transition_mode[k][l];
819 rscale[l] = quant_step_size * s->scale_factor[k][l][sfi] *
820 s->scalefactor_adj[k][sel];
822 if (abits >= 11 || !dca_smpl_bitalloc[abits].vlc[sel].table) {
825 int block_code1, block_code2, size, levels, err;
827 size = abits_sizes[abits - 1];
828 levels = abits_levels[abits - 1];
830 block_code1 = get_bits(&s->gb, size);
831 block_code2 = get_bits(&s->gb, size);
832 err = decode_blockcodes(block_code1, block_code2,
833 levels, block + 8 * l);
835 av_log(s->avctx, AV_LOG_ERROR,
836 "ERROR: block code look-up failed\n");
837 return AVERROR_INVALIDDATA;
841 for (m = 0; m < 8; m++)
842 block[8 * l + m] = get_sbits(&s->gb, abits - 3);
846 for (m = 0; m < 8; m++)
847 block[8 * l + m] = get_bitalloc(&s->gb,
848 &dca_smpl_bitalloc[abits], sel);
853 s->fmt_conv.int32_to_float_fmul_array8(&s->fmt_conv, subband_samples[k][0],
854 block, rscale, 8 * s->vq_start_subband[k]);
856 for (l = 0; l < s->vq_start_subband[k]; l++) {
859 * Inverse ADPCM if in prediction mode
861 if (s->prediction_mode[k][l]) {
863 if (s->predictor_history)
864 subband_samples[k][l][0] += (ff_dca_adpcm_vb[s->prediction_vq[k][l]][0] *
865 s->subband_samples_hist[k][l][3] +
866 ff_dca_adpcm_vb[s->prediction_vq[k][l]][1] *
867 s->subband_samples_hist[k][l][2] +
868 ff_dca_adpcm_vb[s->prediction_vq[k][l]][2] *
869 s->subband_samples_hist[k][l][1] +
870 ff_dca_adpcm_vb[s->prediction_vq[k][l]][3] *
871 s->subband_samples_hist[k][l][0]) *
873 for (m = 1; m < 8; m++) {
874 float sum = ff_dca_adpcm_vb[s->prediction_vq[k][l]][0] *
875 subband_samples[k][l][m - 1];
876 for (n = 2; n <= 4; n++)
878 sum += ff_dca_adpcm_vb[s->prediction_vq[k][l]][n - 1] *
879 subband_samples[k][l][m - n];
880 else if (s->predictor_history)
881 sum += ff_dca_adpcm_vb[s->prediction_vq[k][l]][n - 1] *
882 s->subband_samples_hist[k][l][m - n + 4];
883 subband_samples[k][l][m] += sum * (1.0f / 8192);
889 * Decode VQ encoded high frequencies
891 if (s->subband_activity[k] > s->vq_start_subband[k]) {
892 if (!(s->debug_flag & 0x01)) {
893 av_log(s->avctx, AV_LOG_DEBUG,
894 "Stream with high frequencies VQ coding\n");
895 s->debug_flag |= 0x01;
897 s->dcadsp.decode_hf(subband_samples[k], s->high_freq_vq[k],
898 ff_dca_high_freq_vq, subsubframe * 8,
899 s->scale_factor[k], s->vq_start_subband[k],
900 s->subband_activity[k]);
904 /* Check for DSYNC after subsubframe */
905 if (s->aspf || subsubframe == s->subsubframes[s->current_subframe] - 1) {
906 if (get_bits(&s->gb, 16) != 0xFFFF) {
907 av_log(s->avctx, AV_LOG_ERROR, "Didn't get subframe DSYNC\n");
908 return AVERROR_INVALIDDATA;
912 /* Backup predictor history for adpcm */
913 for (k = base_channel; k < s->prim_channels; k++)
914 for (l = 0; l < s->vq_start_subband[k]; l++)
915 AV_COPY128(s->subband_samples_hist[k][l], &subband_samples[k][l][4]);
920 static int dca_filter_channels(DCAContext *s, int block_index)
922 float (*subband_samples)[DCA_SUBBANDS][8] = s->subband_samples[block_index];
925 /* 32 subbands QMF */
926 for (k = 0; k < s->prim_channels; k++) {
927 if (s->channel_order_tab[k] >= 0)
928 qmf_32_subbands(s, k, subband_samples[k],
929 s->samples_chanptr[s->channel_order_tab[k]],
930 M_SQRT1_2 / 32768.0);
933 /* Generate LFE samples for this subsubframe FIXME!!! */
935 lfe_interpolation_fir(s, s->lfe, 2 * s->lfe,
936 s->lfe_data + 2 * s->lfe * (block_index + 4),
937 s->samples_chanptr[s->lfe_index]);
938 /* Outputs 20bits pcm samples */
941 /* Downmixing to Stereo */
942 if (s->prim_channels + !!s->lfe > 2 &&
943 s->avctx->request_channel_layout == AV_CH_LAYOUT_STEREO) {
944 dca_downmix(s->samples_chanptr, s->amode, !!s->lfe, s->downmix_coef,
945 s->channel_order_tab);
951 static int dca_subframe_footer(DCAContext *s, int base_channel)
953 int in, out, aux_data_count, aux_data_end, reserved;
957 * Unpack optional information
960 /* presumably optional information only appears in the core? */
963 skip_bits_long(&s->gb, 32);
966 aux_data_count = get_bits(&s->gb, 6);
969 skip_bits_long(&s->gb, (-get_bits_count(&s->gb)) & 31);
971 aux_data_end = 8 * aux_data_count + get_bits_count(&s->gb);
973 if ((nsyncaux = get_bits_long(&s->gb, 32)) != DCA_NSYNCAUX) {
974 av_log(s->avctx, AV_LOG_ERROR, "nSYNCAUX mismatch %#"PRIx32"\n",
976 return AVERROR_INVALIDDATA;
979 if (get_bits1(&s->gb)) { // bAUXTimeStampFlag
980 avpriv_request_sample(s->avctx,
981 "Auxiliary Decode Time Stamp Flag");
983 skip_bits(&s->gb, (-get_bits_count(&s->gb)) & 4);
984 // 44 bits: nMSByte (8), nMarker (4), nLSByte (28), nMarker (4)
985 skip_bits_long(&s->gb, 44);
988 if ((s->core_downmix = get_bits1(&s->gb))) {
989 int am = get_bits(&s->gb, 3);
992 s->core_downmix_amode = DCA_MONO;
995 s->core_downmix_amode = DCA_STEREO;
998 s->core_downmix_amode = DCA_STEREO_TOTAL;
1001 s->core_downmix_amode = DCA_3F;
1004 s->core_downmix_amode = DCA_2F1R;
1007 s->core_downmix_amode = DCA_2F2R;
1010 s->core_downmix_amode = DCA_3F1R;
1013 av_log(s->avctx, AV_LOG_ERROR,
1014 "Invalid mode %d for embedded downmix coefficients\n",
1016 return AVERROR_INVALIDDATA;
1018 for (out = 0; out < ff_dca_channels[s->core_downmix_amode]; out++) {
1019 for (in = 0; in < s->prim_channels + !!s->lfe; in++) {
1020 uint16_t tmp = get_bits(&s->gb, 9);
1021 if ((tmp & 0xFF) > 241) {
1022 av_log(s->avctx, AV_LOG_ERROR,
1023 "Invalid downmix coefficient code %"PRIu16"\n",
1025 return AVERROR_INVALIDDATA;
1027 s->core_downmix_codes[in][out] = tmp;
1032 align_get_bits(&s->gb); // byte align
1033 skip_bits(&s->gb, 16); // nAUXCRC16
1035 // additional data (reserved, cf. ETSI TS 102 114 V1.4.1)
1036 if ((reserved = (aux_data_end - get_bits_count(&s->gb))) < 0) {
1037 av_log(s->avctx, AV_LOG_ERROR,
1038 "Overread auxiliary data by %d bits\n", -reserved);
1039 return AVERROR_INVALIDDATA;
1040 } else if (reserved) {
1041 avpriv_request_sample(s->avctx,
1042 "Core auxiliary data reserved content");
1043 skip_bits_long(&s->gb, reserved);
1047 if (s->crc_present && s->dynrange)
1048 get_bits(&s->gb, 16);
1055 * Decode a dca frame block
1057 * @param s pointer to the DCAContext
1060 static int dca_decode_block(DCAContext *s, int base_channel, int block_index)
1065 if (s->current_subframe >= s->subframes) {
1066 av_log(s->avctx, AV_LOG_DEBUG, "check failed: %i>%i",
1067 s->current_subframe, s->subframes);
1068 return AVERROR_INVALIDDATA;
1071 if (!s->current_subsubframe) {
1072 /* Read subframe header */
1073 if ((ret = dca_subframe_header(s, base_channel, block_index)))
1077 /* Read subsubframe */
1078 if ((ret = dca_subsubframe(s, base_channel, block_index)))
1082 s->current_subsubframe++;
1083 if (s->current_subsubframe >= s->subsubframes[s->current_subframe]) {
1084 s->current_subsubframe = 0;
1085 s->current_subframe++;
1087 if (s->current_subframe >= s->subframes) {
1088 /* Read subframe footer */
1089 if ((ret = dca_subframe_footer(s, base_channel)))
1096 int ff_dca_xbr_parse_frame(DCAContext *s)
1098 int scale_table_high[DCA_CHSET_CHANS_MAX][DCA_SUBBANDS][2];
1099 int active_bands[DCA_CHSETS_MAX][DCA_CHSET_CHANS_MAX];
1100 int abits_high[DCA_CHSET_CHANS_MAX][DCA_SUBBANDS];
1101 int anctemp[DCA_CHSET_CHANS_MAX];
1102 int chset_fsize[DCA_CHSETS_MAX];
1103 int n_xbr_ch[DCA_CHSETS_MAX];
1104 int hdr_size, num_chsets, xbr_tmode, hdr_pos;
1105 int i, j, k, l, chset, chan_base;
1107 av_log(s->avctx, AV_LOG_DEBUG, "DTS-XBR: decoding XBR extension\n");
1109 /* get bit position of sync header */
1110 hdr_pos = get_bits_count(&s->gb) - 32;
1112 hdr_size = get_bits(&s->gb, 6) + 1;
1113 num_chsets = get_bits(&s->gb, 2) + 1;
1115 for(i = 0; i < num_chsets; i++)
1116 chset_fsize[i] = get_bits(&s->gb, 14) + 1;
1118 xbr_tmode = get_bits1(&s->gb);
1120 for(i = 0; i < num_chsets; i++) {
1121 n_xbr_ch[i] = get_bits(&s->gb, 3) + 1;
1122 k = get_bits(&s->gb, 2) + 5;
1123 for(j = 0; j < n_xbr_ch[i]; j++)
1124 active_bands[i][j] = get_bits(&s->gb, k) + 1;
1127 /* skip to the end of the header */
1128 i = get_bits_count(&s->gb);
1129 if(hdr_pos + hdr_size * 8 > i)
1130 skip_bits_long(&s->gb, hdr_pos + hdr_size * 8 - i);
1132 /* loop over the channel data sets */
1133 /* only decode as many channels as we've decoded base data for */
1134 for(chset = 0, chan_base = 0;
1135 chset < num_chsets && chan_base + n_xbr_ch[chset] <= s->prim_channels;
1136 chan_base += n_xbr_ch[chset++]) {
1137 int start_posn = get_bits_count(&s->gb);
1138 int subsubframe = 0;
1141 /* loop over subframes */
1142 for (k = 0; k < (s->sample_blocks / 8); k++) {
1143 /* parse header if we're on first subsubframe of a block */
1144 if(subsubframe == 0) {
1145 /* Parse subframe header */
1146 for(i = 0; i < n_xbr_ch[chset]; i++) {
1147 anctemp[i] = get_bits(&s->gb, 2) + 2;
1150 for(i = 0; i < n_xbr_ch[chset]; i++) {
1151 get_array(&s->gb, abits_high[i], active_bands[chset][i], anctemp[i]);
1154 for(i = 0; i < n_xbr_ch[chset]; i++) {
1155 anctemp[i] = get_bits(&s->gb, 3);
1156 if(anctemp[i] < 1) {
1157 av_log(s->avctx, AV_LOG_ERROR, "DTS-XBR: SYNC ERROR\n");
1158 return AVERROR_INVALIDDATA;
1162 /* generate scale factors */
1163 for(i = 0; i < n_xbr_ch[chset]; i++) {
1164 const uint32_t *scale_table;
1167 if (s->scalefactor_huffman[chan_base+i] == 6) {
1168 scale_table = ff_dca_scale_factor_quant7;
1170 scale_table = ff_dca_scale_factor_quant6;
1175 for(j = 0; j < active_bands[chset][i]; j++) {
1176 if(abits_high[i][j] > 0) {
1177 scale_table_high[i][j][0] =
1178 scale_table[get_bits(&s->gb, nbits)];
1180 if(xbr_tmode && s->transition_mode[i][j]) {
1181 scale_table_high[i][j][1] =
1182 scale_table[get_bits(&s->gb, nbits)];
1189 /* decode audio array for this block */
1190 for(i = 0; i < n_xbr_ch[chset]; i++) {
1191 for(j = 0; j < active_bands[chset][i]; j++) {
1192 const int xbr_abits = abits_high[i][j];
1193 const float quant_step_size = ff_dca_lossless_quant_d[xbr_abits];
1194 const int sfi = xbr_tmode && s->transition_mode[i][j] && subsubframe >= s->transition_mode[i][j];
1195 const float rscale = quant_step_size * scale_table_high[i][j][sfi];
1196 float *subband_samples = s->subband_samples[k][chan_base+i][j];
1203 get_array(&s->gb, block, 8, xbr_abits - 3);
1205 int block_code1, block_code2, size, levels, err;
1207 size = abits_sizes[xbr_abits - 1];
1208 levels = abits_levels[xbr_abits - 1];
1210 block_code1 = get_bits(&s->gb, size);
1211 block_code2 = get_bits(&s->gb, size);
1212 err = decode_blockcodes(block_code1, block_code2,
1215 av_log(s->avctx, AV_LOG_ERROR,
1216 "ERROR: DTS-XBR: block code look-up failed\n");
1217 return AVERROR_INVALIDDATA;
1221 /* scale & sum into subband */
1222 for(l = 0; l < 8; l++)
1223 subband_samples[l] += (float)block[l] * rscale;
1227 /* check DSYNC marker */
1228 if(s->aspf || subsubframe == s->subsubframes[subframe] - 1) {
1229 if(get_bits(&s->gb, 16) != 0xffff) {
1230 av_log(s->avctx, AV_LOG_ERROR, "DTS-XBR: Didn't get subframe DSYNC\n");
1231 return AVERROR_INVALIDDATA;
1235 /* advance sub-sub-frame index */
1236 if(++subsubframe >= s->subsubframes[subframe]) {
1242 /* skip to next channel set */
1243 i = get_bits_count(&s->gb);
1244 if(start_posn + chset_fsize[chset] * 8 != i) {
1245 j = start_posn + chset_fsize[chset] * 8 - i;
1247 av_log(s->avctx, AV_LOG_ERROR, "DTS-XBR: end of channel set,"
1248 " skipping further than expected (%d bits)\n", j);
1249 skip_bits_long(&s->gb, j);
1257 /* parse initial header for XXCH and dump details */
1258 int ff_dca_xxch_decode_frame(DCAContext *s)
1260 int hdr_size, spkmsk_bits, num_chsets, core_spk, hdr_pos;
1261 int i, chset, base_channel, chstart, fsize[8];
1263 /* assume header word has already been parsed */
1264 hdr_pos = get_bits_count(&s->gb) - 32;
1265 hdr_size = get_bits(&s->gb, 6) + 1;
1266 /*chhdr_crc =*/ skip_bits1(&s->gb);
1267 spkmsk_bits = get_bits(&s->gb, 5) + 1;
1268 num_chsets = get_bits(&s->gb, 2) + 1;
1270 for (i = 0; i < num_chsets; i++)
1271 fsize[i] = get_bits(&s->gb, 14) + 1;
1273 core_spk = get_bits(&s->gb, spkmsk_bits);
1274 s->xxch_core_spkmask = core_spk;
1275 s->xxch_nbits_spk_mask = spkmsk_bits;
1276 s->xxch_dmix_embedded = 0;
1278 /* skip to the end of the header */
1279 i = get_bits_count(&s->gb);
1280 if (hdr_pos + hdr_size * 8 > i)
1281 skip_bits_long(&s->gb, hdr_pos + hdr_size * 8 - i);
1283 for (chset = 0; chset < num_chsets; chset++) {
1284 chstart = get_bits_count(&s->gb);
1285 base_channel = s->prim_channels;
1286 s->xxch_chset = chset;
1288 /* XXCH and Core headers differ, see 6.4.2 "XXCH Channel Set Header" vs.
1289 5.3.2 "Primary Audio Coding Header", DTS Spec 1.3.1 */
1290 dca_parse_audio_coding_header(s, base_channel, 1);
1292 /* decode channel data */
1293 for (i = 0; i < (s->sample_blocks / 8); i++) {
1294 if (dca_decode_block(s, base_channel, i)) {
1295 av_log(s->avctx, AV_LOG_ERROR,
1296 "Error decoding DTS-XXCH extension\n");
1301 /* skip to end of this section */
1302 i = get_bits_count(&s->gb);
1303 if (chstart + fsize[chset] * 8 > i)
1304 skip_bits_long(&s->gb, chstart + fsize[chset] * 8 - i);
1306 s->xxch_chset = num_chsets;
1311 static float dca_dmix_code(unsigned code)
1313 int sign = (code >> 8) - 1;
1315 return ((ff_dca_dmixtable[code] ^ sign) - sign) * (1.0 / (1 << 15));
1319 * Main frame decoding function
1320 * FIXME add arguments
1322 static int dca_decode_frame(AVCodecContext *avctx, void *data,
1323 int *got_frame_ptr, AVPacket *avpkt)
1325 AVFrame *frame = data;
1326 const uint8_t *buf = avpkt->data;
1327 int buf_size = avpkt->size;
1331 int num_core_channels = 0;
1333 float **samples_flt;
1336 DCAContext *s = avctx->priv_data;
1338 int channels, full_channels;
1350 s->dca_buffer_size = avpriv_dca_convert_bitstream(buf, buf_size, s->dca_buffer,
1351 DCA_MAX_FRAME_SIZE + DCA_MAX_EXSS_HEADER_SIZE);
1352 if (s->dca_buffer_size == AVERROR_INVALIDDATA) {
1353 av_log(avctx, AV_LOG_ERROR, "Not a valid DCA frame\n");
1354 return AVERROR_INVALIDDATA;
1357 if ((ret = dca_parse_frame_header(s)) < 0) {
1358 // seems like the frame is corrupt, try with the next one
1361 // set AVCodec values with parsed data
1362 avctx->sample_rate = s->sample_rate;
1363 avctx->bit_rate = s->bit_rate;
1365 s->profile = FF_PROFILE_DTS;
1367 for (i = 0; i < (s->sample_blocks / 8); i++) {
1368 if ((ret = dca_decode_block(s, 0, i))) {
1369 av_log(avctx, AV_LOG_ERROR, "error decoding block\n");
1374 /* record number of core channels incase less than max channels are requested */
1375 num_core_channels = s->prim_channels;
1377 if (s->prim_channels + !!s->lfe > 2 &&
1378 avctx->request_channel_layout == AV_CH_LAYOUT_STEREO) {
1379 /* Stereo downmix coefficients
1381 * The decoder can only downmix to 2-channel, so we need to ensure
1382 * embedded downmix coefficients are actually targeting 2-channel.
1384 if (s->core_downmix && (s->core_downmix_amode == DCA_STEREO ||
1385 s->core_downmix_amode == DCA_STEREO_TOTAL)) {
1386 for (i = 0; i < num_core_channels + !!s->lfe; i++) {
1387 /* Range checked earlier */
1388 s->downmix_coef[i][0] = dca_dmix_code(s->core_downmix_codes[i][0]);
1389 s->downmix_coef[i][1] = dca_dmix_code(s->core_downmix_codes[i][1]);
1391 s->output = s->core_downmix_amode;
1393 int am = s->amode & DCA_CHANNEL_MASK;
1394 if (am >= FF_ARRAY_ELEMS(ff_dca_default_coeffs)) {
1395 av_log(s->avctx, AV_LOG_ERROR,
1396 "Invalid channel mode %d\n", am);
1397 return AVERROR_INVALIDDATA;
1399 if (num_core_channels + !!s->lfe >
1400 FF_ARRAY_ELEMS(ff_dca_default_coeffs[0])) {
1401 avpriv_request_sample(s->avctx, "Downmixing %d channels",
1402 s->prim_channels + !!s->lfe);
1403 return AVERROR_PATCHWELCOME;
1405 for (i = 0; i < num_core_channels + !!s->lfe; i++) {
1406 s->downmix_coef[i][0] = ff_dca_default_coeffs[am][i][0];
1407 s->downmix_coef[i][1] = ff_dca_default_coeffs[am][i][1];
1410 av_dlog(s->avctx, "Stereo downmix coeffs:\n");
1411 for (i = 0; i < num_core_channels + !!s->lfe; i++) {
1412 av_dlog(s->avctx, "L, input channel %d = %f\n", i,
1413 s->downmix_coef[i][0]);
1414 av_dlog(s->avctx, "R, input channel %d = %f\n", i,
1415 s->downmix_coef[i][1]);
1417 av_dlog(s->avctx, "\n");
1421 s->core_ext_mask = ff_dca_ext_audio_descr_mask[s->ext_descr];
1423 s->core_ext_mask = 0;
1425 core_ss_end = FFMIN(s->frame_size, s->dca_buffer_size) * 8;
1427 /* only scan for extensions if ext_descr was unknown or indicated a
1428 * supported XCh extension */
1429 if (s->core_ext_mask < 0 || s->core_ext_mask & (DCA_EXT_XCH | DCA_EXT_XXCH)) {
1430 /* if ext_descr was unknown, clear s->core_ext_mask so that the
1431 * extensions scan can fill it up */
1432 s->core_ext_mask = FFMAX(s->core_ext_mask, 0);
1434 /* extensions start at 32-bit boundaries into bitstream */
1435 skip_bits_long(&s->gb, (-get_bits_count(&s->gb)) & 31);
1437 while (core_ss_end - get_bits_count(&s->gb) >= 32) {
1438 uint32_t bits = get_bits_long(&s->gb, 32);
1441 case DCA_SYNCWORD_XCH: {
1442 int ext_amode, xch_fsize;
1444 s->xch_base_channel = s->prim_channels;
1446 /* validate sync word using XCHFSIZE field */
1447 xch_fsize = show_bits(&s->gb, 10);
1448 if ((s->frame_size != (get_bits_count(&s->gb) >> 3) - 4 + xch_fsize) &&
1449 (s->frame_size != (get_bits_count(&s->gb) >> 3) - 4 + xch_fsize + 1))
1452 /* skip length-to-end-of-frame field for the moment */
1453 skip_bits(&s->gb, 10);
1455 s->core_ext_mask |= DCA_EXT_XCH;
1457 /* extension amode(number of channels in extension) should be 1 */
1458 /* AFAIK XCh is not used for more channels */
1459 if ((ext_amode = get_bits(&s->gb, 4)) != 1) {
1460 av_log(avctx, AV_LOG_ERROR,
1461 "XCh extension amode %d not supported!\n",
1466 if (s->xch_base_channel < 2) {
1467 avpriv_request_sample(avctx, "XCh with fewer than 2 base channels");
1471 /* much like core primary audio coding header */
1472 dca_parse_audio_coding_header(s, s->xch_base_channel, 0);
1474 for (i = 0; i < (s->sample_blocks / 8); i++)
1475 if ((ret = dca_decode_block(s, s->xch_base_channel, i))) {
1476 av_log(avctx, AV_LOG_ERROR, "error decoding XCh extension\n");
1483 case DCA_SYNCWORD_XXCH:
1484 /* XXCh: extended channels */
1485 /* usually found either in core or HD part in DTS-HD HRA streams,
1486 * but not in DTS-ES which contains XCh extensions instead */
1487 s->core_ext_mask |= DCA_EXT_XXCH;
1488 ff_dca_xxch_decode_frame(s);
1492 int fsize96 = show_bits(&s->gb, 12) + 1;
1493 if (s->frame_size != (get_bits_count(&s->gb) >> 3) - 4 + fsize96)
1496 av_log(avctx, AV_LOG_DEBUG, "X96 extension found at %d bits\n",
1497 get_bits_count(&s->gb));
1498 skip_bits(&s->gb, 12);
1499 av_log(avctx, AV_LOG_DEBUG, "FSIZE96 = %d bytes\n", fsize96);
1500 av_log(avctx, AV_LOG_DEBUG, "REVNO = %d\n", get_bits(&s->gb, 4));
1502 s->core_ext_mask |= DCA_EXT_X96;
1507 skip_bits_long(&s->gb, (-get_bits_count(&s->gb)) & 31);
1510 /* no supported extensions, skip the rest of the core substream */
1511 skip_bits_long(&s->gb, core_ss_end - get_bits_count(&s->gb));
1514 if (s->core_ext_mask & DCA_EXT_X96)
1515 s->profile = FF_PROFILE_DTS_96_24;
1516 else if (s->core_ext_mask & (DCA_EXT_XCH | DCA_EXT_XXCH))
1517 s->profile = FF_PROFILE_DTS_ES;
1519 /* check for ExSS (HD part) */
1520 if (s->dca_buffer_size - s->frame_size > 32 &&
1521 get_bits_long(&s->gb, 32) == DCA_SYNCWORD_SUBSTREAM)
1522 ff_dca_exss_parse_header(s);
1524 avctx->profile = s->profile;
1526 full_channels = channels = s->prim_channels + !!s->lfe;
1528 /* If we have XXCH then the channel layout is managed differently */
1529 /* note that XLL will also have another way to do things */
1530 if (!(s->core_ext_mask & DCA_EXT_XXCH)
1531 || (s->core_ext_mask & DCA_EXT_XXCH && avctx->request_channels > 0
1532 && avctx->request_channels
1533 < num_core_channels + !!s->lfe + s->xxch_chset_nch[0]))
1534 { /* xxx should also do MA extensions */
1535 if (s->amode < 16) {
1536 avctx->channel_layout = ff_dca_core_channel_layout[s->amode];
1538 if (s->prim_channels + !!s->lfe > 2 &&
1539 avctx->request_channel_layout == AV_CH_LAYOUT_STEREO) {
1541 * Neither the core's auxiliary data nor our default tables contain
1542 * downmix coefficients for the additional channel coded in the XCh
1543 * extension, so when we're doing a Stereo downmix, don't decode it.
1548 #if FF_API_REQUEST_CHANNELS
1549 FF_DISABLE_DEPRECATION_WARNINGS
1550 if (s->xch_present && !s->xch_disable &&
1551 (!avctx->request_channels ||
1552 avctx->request_channels > num_core_channels + !!s->lfe)) {
1553 FF_ENABLE_DEPRECATION_WARNINGS
1555 if (s->xch_present && !s->xch_disable) {
1557 if (avctx->channel_layout & AV_CH_BACK_CENTER) {
1558 avpriv_request_sample(avctx, "XCh with Back center channel");
1559 return AVERROR_INVALIDDATA;
1561 avctx->channel_layout |= AV_CH_BACK_CENTER;
1563 avctx->channel_layout |= AV_CH_LOW_FREQUENCY;
1564 s->channel_order_tab = ff_dca_channel_reorder_lfe_xch[s->amode];
1566 s->channel_order_tab = ff_dca_channel_reorder_nolfe_xch[s->amode];
1568 if (s->channel_order_tab[s->xch_base_channel] < 0)
1569 return AVERROR_INVALIDDATA;
1571 channels = num_core_channels + !!s->lfe;
1572 s->xch_present = 0; /* disable further xch processing */
1574 avctx->channel_layout |= AV_CH_LOW_FREQUENCY;
1575 s->channel_order_tab = ff_dca_channel_reorder_lfe[s->amode];
1577 s->channel_order_tab = ff_dca_channel_reorder_nolfe[s->amode];
1580 if (channels > !!s->lfe &&
1581 s->channel_order_tab[channels - 1 - !!s->lfe] < 0)
1582 return AVERROR_INVALIDDATA;
1584 if (av_get_channel_layout_nb_channels(avctx->channel_layout) != channels) {
1585 av_log(avctx, AV_LOG_ERROR, "Number of channels %d mismatches layout %d\n", channels, av_get_channel_layout_nb_channels(avctx->channel_layout));
1586 return AVERROR_INVALIDDATA;
1589 if (num_core_channels + !!s->lfe > 2 &&
1590 avctx->request_channel_layout == AV_CH_LAYOUT_STEREO) {
1592 s->output = s->prim_channels == 2 ? s->amode : DCA_STEREO;
1593 avctx->channel_layout = AV_CH_LAYOUT_STEREO;
1595 else if (avctx->request_channel_layout & AV_CH_LAYOUT_NATIVE) {
1596 static const int8_t dca_channel_order_native[9] = { 0, 1, 2, 3, 4, 5, 6, 7, 8 };
1597 s->channel_order_tab = dca_channel_order_native;
1599 s->lfe_index = ff_dca_lfe_index[s->amode];
1601 av_log(avctx, AV_LOG_ERROR,
1602 "Non standard configuration %d !\n", s->amode);
1603 return AVERROR_INVALIDDATA;
1606 s->xxch_dmix_embedded = 0;
1608 /* we only get here if an XXCH channel set can be added to the mix */
1609 channel_mask = s->xxch_core_spkmask;
1611 if (avctx->request_channels > 0
1612 && avctx->request_channels < s->prim_channels) {
1613 channels = num_core_channels + !!s->lfe;
1614 for (i = 0; i < s->xxch_chset && channels + s->xxch_chset_nch[i]
1615 <= avctx->request_channels; i++) {
1616 channels += s->xxch_chset_nch[i];
1617 channel_mask |= s->xxch_spk_masks[i];
1620 channels = s->prim_channels + !!s->lfe;
1621 for (i = 0; i < s->xxch_chset; i++) {
1622 channel_mask |= s->xxch_spk_masks[i];
1626 /* Given the DTS spec'ed channel mask, generate an avcodec version */
1628 for (i = 0; i < s->xxch_nbits_spk_mask; ++i) {
1629 if (channel_mask & (1 << i)) {
1630 channel_layout |= ff_dca_map_xxch_to_native[i];
1634 /* make sure that we have managed to get equivalent dts/avcodec channel
1635 * masks in some sense -- unfortunately some channels could overlap */
1636 if (av_popcount(channel_mask) != av_popcount(channel_layout)) {
1637 av_log(avctx, AV_LOG_DEBUG,
1638 "DTS-XXCH: Inconsistent avcodec/dts channel layouts\n");
1639 return AVERROR_INVALIDDATA;
1642 avctx->channel_layout = channel_layout;
1644 if (!(avctx->request_channel_layout & AV_CH_LAYOUT_NATIVE)) {
1645 /* Estimate DTS --> avcodec ordering table */
1646 for (chset = -1, j = 0; chset < s->xxch_chset; ++chset) {
1647 mask = chset >= 0 ? s->xxch_spk_masks[chset]
1648 : s->xxch_core_spkmask;
1649 for (i = 0; i < s->xxch_nbits_spk_mask; i++) {
1650 if (mask & ~(DCA_XXCH_LFE1 | DCA_XXCH_LFE2) & (1 << i)) {
1651 lavc = ff_dca_map_xxch_to_native[i];
1652 posn = av_popcount(channel_layout & (lavc - 1));
1653 s->xxch_order_tab[j++] = posn;
1659 s->lfe_index = av_popcount(channel_layout & (AV_CH_LOW_FREQUENCY-1));
1660 } else { /* native ordering */
1661 for (i = 0; i < channels; i++)
1662 s->xxch_order_tab[i] = i;
1664 s->lfe_index = channels - 1;
1667 s->channel_order_tab = s->xxch_order_tab;
1670 if (avctx->channels != channels) {
1671 if (avctx->channels)
1672 av_log(avctx, AV_LOG_INFO, "Number of channels changed in DCA decoder (%d -> %d)\n", avctx->channels, channels);
1673 avctx->channels = channels;
1676 /* get output buffer */
1677 frame->nb_samples = 256 * (s->sample_blocks / 8);
1678 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1680 samples_flt = (float **) frame->extended_data;
1682 /* allocate buffer for extra channels if downmixing */
1683 if (avctx->channels < full_channels) {
1684 ret = av_samples_get_buffer_size(NULL, full_channels - channels,
1686 avctx->sample_fmt, 0);
1690 av_fast_malloc(&s->extra_channels_buffer,
1691 &s->extra_channels_buffer_size, ret);
1692 if (!s->extra_channels_buffer)
1693 return AVERROR(ENOMEM);
1695 ret = av_samples_fill_arrays((uint8_t **) s->extra_channels, NULL,
1696 s->extra_channels_buffer,
1697 full_channels - channels,
1698 frame->nb_samples, avctx->sample_fmt, 0);
1703 /* filter to get final output */
1704 for (i = 0; i < (s->sample_blocks / 8); i++) {
1707 for (ch = 0; ch < channels; ch++)
1708 s->samples_chanptr[ch] = samples_flt[ch] + i * 256;
1709 for (; ch < full_channels; ch++)
1710 s->samples_chanptr[ch] = s->extra_channels[ch - channels] + i * 256;
1712 dca_filter_channels(s, i);
1714 /* If this was marked as a DTS-ES stream we need to subtract back- */
1715 /* channel from SL & SR to remove matrixed back-channel signal */
1716 if ((s->source_pcm_res & 1) && s->xch_present) {
1717 float *back_chan = s->samples_chanptr[s->channel_order_tab[s->xch_base_channel]];
1718 float *lt_chan = s->samples_chanptr[s->channel_order_tab[s->xch_base_channel - 2]];
1719 float *rt_chan = s->samples_chanptr[s->channel_order_tab[s->xch_base_channel - 1]];
1720 s->fdsp->vector_fmac_scalar(lt_chan, back_chan, -M_SQRT1_2, 256);
1721 s->fdsp->vector_fmac_scalar(rt_chan, back_chan, -M_SQRT1_2, 256);
1724 /* If stream contains XXCH, we might need to undo an embedded downmix */
1725 if (s->xxch_dmix_embedded) {
1726 /* Loop over channel sets in turn */
1727 ch = num_core_channels;
1728 for (chset = 0; chset < s->xxch_chset; chset++) {
1729 endch = ch + s->xxch_chset_nch[chset];
1730 mask = s->xxch_dmix_embedded;
1733 for (j = ch; j < endch; j++) {
1734 if (mask & (1 << j)) { /* this channel has been mixed-out */
1735 src_chan = s->samples_chanptr[s->channel_order_tab[j]];
1736 for (k = 0; k < endch; k++) {
1737 achan = s->channel_order_tab[k];
1738 scale = s->xxch_dmix_coeff[j][k];
1740 dst_chan = s->samples_chanptr[achan];
1741 s->fdsp->vector_fmac_scalar(dst_chan, src_chan,
1748 /* if a downmix has been embedded then undo the pre-scaling */
1749 if ((mask & (1 << ch)) && s->xxch_dmix_sf[chset] != 1.0f) {
1750 scale = s->xxch_dmix_sf[chset];
1752 for (j = 0; j < ch; j++) {
1753 src_chan = s->samples_chanptr[s->channel_order_tab[j]];
1754 for (k = 0; k < 256; k++)
1755 src_chan[k] *= scale;
1758 /* LFE channel is always part of core, scale if it exists */
1760 src_chan = s->samples_chanptr[s->lfe_index];
1761 for (k = 0; k < 256; k++)
1762 src_chan[k] *= scale;
1772 /* update lfe history */
1773 lfe_samples = 2 * s->lfe * (s->sample_blocks / 8);
1774 for (i = 0; i < 2 * s->lfe * 4; i++)
1775 s->lfe_data[i] = s->lfe_data[i + lfe_samples];
1779 * DCA_STEREO_TOTAL (Lt/Rt) is equivalent to Dolby Surround */
1780 ret = ff_side_data_update_matrix_encoding(frame,
1781 (s->output & ~DCA_LFE) == DCA_STEREO_TOTAL ?
1782 AV_MATRIX_ENCODING_DOLBY : AV_MATRIX_ENCODING_NONE);
1792 * DCA initialization
1794 * @param avctx pointer to the AVCodecContext
1797 static av_cold int dca_decode_init(AVCodecContext *avctx)
1799 DCAContext *s = avctx->priv_data;
1804 s->fdsp = avpriv_float_dsp_alloc(avctx->flags & CODEC_FLAG_BITEXACT);
1806 return AVERROR(ENOMEM);
1808 ff_mdct_init(&s->imdct, 6, 1, 1.0);
1809 ff_synth_filter_init(&s->synth);
1810 ff_dcadsp_init(&s->dcadsp);
1811 ff_fmt_convert_init(&s->fmt_conv, avctx);
1813 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
1815 /* allow downmixing to stereo */
1816 #if FF_API_REQUEST_CHANNELS
1817 FF_DISABLE_DEPRECATION_WARNINGS
1818 if (avctx->request_channels == 2)
1819 avctx->request_channel_layout = AV_CH_LAYOUT_STEREO;
1820 FF_ENABLE_DEPRECATION_WARNINGS
1822 if (avctx->channels > 2 &&
1823 avctx->request_channel_layout == AV_CH_LAYOUT_STEREO)
1824 avctx->channels = 2;
1829 static av_cold int dca_decode_end(AVCodecContext *avctx)
1831 DCAContext *s = avctx->priv_data;
1832 ff_mdct_end(&s->imdct);
1833 av_freep(&s->extra_channels_buffer);
1838 static const AVProfile profiles[] = {
1839 { FF_PROFILE_DTS, "DTS" },
1840 { FF_PROFILE_DTS_ES, "DTS-ES" },
1841 { FF_PROFILE_DTS_96_24, "DTS 96/24" },
1842 { FF_PROFILE_DTS_HD_HRA, "DTS-HD HRA" },
1843 { FF_PROFILE_DTS_HD_MA, "DTS-HD MA" },
1844 { FF_PROFILE_UNKNOWN },
1847 static const AVOption options[] = {
1848 { "disable_xch", "disable decoding of the XCh extension", offsetof(DCAContext, xch_disable), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM },
1852 static const AVClass dca_decoder_class = {
1853 .class_name = "DCA decoder",
1854 .item_name = av_default_item_name,
1856 .version = LIBAVUTIL_VERSION_INT,
1857 .category = AV_CLASS_CATEGORY_DECODER,
1860 AVCodec ff_dca_decoder = {
1862 .long_name = NULL_IF_CONFIG_SMALL("DCA (DTS Coherent Acoustics)"),
1863 .type = AVMEDIA_TYPE_AUDIO,
1864 .id = AV_CODEC_ID_DTS,
1865 .priv_data_size = sizeof(DCAContext),
1866 .init = dca_decode_init,
1867 .decode = dca_decode_frame,
1868 .close = dca_decode_end,
1869 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
1870 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1871 AV_SAMPLE_FMT_NONE },
1872 .profiles = NULL_IF_CONFIG_SMALL(profiles),
1873 .priv_class = &dca_decoder_class,